This is a talk given at the Redwood Center for Theoretical Neuroscience, UC Berkeley on November 11, 2006 by Urs Köster from the University of Helsinki.
Since the discovery of Simple and Complex cells in primary visual cortex, which have easily visualized receptive fields, efforts have been made to build models that show the emergence of similar features using unsupervised learning techniques applied to natural image data. Here we discuss recent work on extending single layer models to give rise to complex cell receptive fields. We proceed to describe a novel method called “Score Matching” for the estimation of multi-layer statistical models and the application to modeling of primary visual cortex.
Kilian Koepsell Redwood Center for Theoretical Neuroscience University of California Helen Wills Neuroscience Institute 132 Barker, MC #3190 Berkeley, CA 94720-3190